OpenArt published these six images when it added Nano Banana 2.1 to its model catalog. They show what the model is pitched for: a product line, annotated packaging, a type-heavy menu board, a full brand landing page, a photo restyled from a reference, and a large soft scene. The prompts on the cards are new prompt ideas written by this site; OpenArt did not publish the prompts that made the images.
Third-party samples published by OpenArt to present Google's Nano Banana 2.1, captured from its model page and launch clip on 7 October 2026. None was generated on this site.
A product family photograph of a complete skincare line on a seamless sky-blue backdrop: eight white and frosted-glass bottles and jars with minimal navy typography reading ELUNEA and DEW RITUAL, a dropper mid-drip above an open serum, two dark grapes resting on the glossy reflective counter, soft diffused daylight, a clear reflection under every item, high-end cosmetics campaign finish.
A clean e-commerce packaging photograph: four white pump bottles and a body scrub jar floating at slight angles on a near-white background, minimal purple and blue typography naming each product, small rounded callout labels with thin leader lines pointing at the shampoo, the shower gel, the body scrub and the cleansing foam, soft studio shadow, generous negative space.
A landscape cafe menu board in flat graphic style on a pale blue background: four dish photographs inside white circles, the venue name in oversized mixed black and blue lettering, four columns listing dishes with short ingredient and preparation notes, a row of rounded navigation labels along the top, clean vector composition, a print-ready poster.
A full-width landing page design for a furniture brand called MELLWFORM: deep blue header with an oversized cream wordmark, a hero panel showing a translucent orange armchair beside the headline Furniture with personality, four product series cards below with photos, names and prices, rounded buttons, a Swiss-style grid, flat web design mockup.
Redraw the reference portrait as a clean flat illustration: a young woman in a navy knit beanie and pink buttoned shirt holding a lily stem in her teeth, oversized white and pink lily blooms and long green leaves framing her face, smooth cel-style shading, minimal outlines, a soft matte palette, editorial illustration finish.
A wide cinematic photograph: a giant fluffy pink teddy bear and a low grey fabric sofa standing side by side in an overgrown garden of oversized lilies, mushrooms and wildflowers, a girl in cream loungewear resting against the bear, a pink-clouded sky, moss underfoot, a dreamy pastel palette, high-detail editorial photography.
New prompt idea; not the source image's own prompt.
What Nano Banana 2.1 is
Nano Banana 2.1 is Google's image generation and editing model, released as a generally available version on 6 October 2026 under the model ID gemini-nano-banana-2.1. Google's own documentation describes it as optimized for multimodal image generation and editing, positioned in the Flash tier as the successor to Nano Banana 2, while Nano Banana Pro remains the family's higher-fidelity tier.
On this site it runs in the same image workspace as every other model, with text-to-image and image-to-image as the modes this service actually opens. This page documents the model, the controls this workspace exposes, starter prompts you can copy, and the quote the service returns before a task is submitted.
Jobs Nano Banana 2.1 is suited to
The list follows what Google documents for this model and what this service's catalog confirms, not what resellers claim about it.
Marketing visuals and posters with legible lettering. Name the headline in quotation marks and say where it sits.
Product photography and variations. Describe the change in plain language and state what must stay unchanged.
Reference-based editing without masks. Upload the images and describe the change; the model keeps the rest.
Infographics and labelled diagrams, a documented strength of this model family.
Virtual try-on, which Google lists as a supported capability for this model.
Grounded visuals. Google documents Google Search and image search grounding for this model.
Multi-turn editing, refining the same image across several instructions in sequence.
Interleaved image and text output, which Google documents as supported.
Images derived from video input, documented by Google though not exposed by this service.
Starter prompts you can copy
The prompts below are written by this site as starting points for the task types this model family is known for. They are not publisher quotations and they are not results this site has run. Treat them as test briefs, keep the structure, and replace the subject with your own.
Poster with a headline
A vertical concert poster for an indie folk evening, warm cream paper background, hand-lettered headline Folk Night at the Harbor in quotation marks across the top, small date and venue line below it, soft ink texture, no photorealistic people
Product in a scene
A studio product photograph of a matte ceramic pour-over coffee dripper on a walnut counter, morning light from the left, shallow depth of field, a small steaming cup beside it, clean minimal styling
Reference edit
Take the uploaded photo and change only the background to a bright modern kitchen, keep the product, its angle, its label and its lighting exactly as they are
Text layout
A bakery menu card, off-white background, elegant serif title Seasonal Menu centered at the top, five item lines with dotted leaders and prices, restrained two-color palette, print-ready composition
Character consistency
Using the uploaded portrait as reference, show the same person in three settings: a library, a rainy street at night, and a sunny rooftop, keeping the face, hairstyle and outfit consistent across all three images
Specifications Google publishes
These figures come from Google's own documentation for this model, retrieved on 7 October 2026, not from a reseller's summary.
Item
Google's published figure
Model ID
gemini-nano-banana-2.1
Release status
Generally available, 6 October 2026
Context window
131,072 tokens
Input images per prompt
Up to 14
Input file size
7 MB inline or in the console, 30 MB via Cloud Storage
Sampling controls
seed, temperature, topP and topK are not supported and are rejected
Person generation
Not supported
Audio
Not supported
Grounding
Google Search and image search
Provenance
C2PA Content Credentials
Other documented capabilities
Multi-turn editing, virtual try-on, interleaved image and text output, thinking mode
Output resolutions
1K, 2K and 4K
What this workspace exposes
The picker is the authority on what you can submit here. The figures below come from the live Star Task catalog retrieved on 7 October 2026; another provider's page describes that provider's account, not this one, and the service resolves the price rather than this page.
For scale, fal lists 0.08 USD per image at 1K with 2K billed at 1.5 times and 4K at twice the base rate, and OpenRouter lists 1.50 USD per million input tokens and 30.00 USD per million output tokens. Both are their own rate cards for their own accounts, quoted here only to show how third-party pricing is usually structured. Your own number is the one the workspace prints.
Control
Options in this workspace
Modes
Text-to-image and image-to-image
Resolution tier
1K (default), 2K, 4K
Aspect ratio
14 fixed values from 1:1 to 21:9, including 9:16, 16:9, 4:3 and the extremes 1:4, 4:1, 1:8 and 8:1; no auto value
Reference images
Up to 14 in one request
Cost
From 3 credits per result at 1K, 4 credits at 2K and 6 credits at 4K, resolved by the service; the quote shown before submission is the binding figure
Nano Banana 2.1, 2 and 2 Lite compared
All three tiers are listed on this site. The positioning rows are Google's; the control rows are this workspace's own picker, which is why they can differ from what another site offers.
Nano Banana Pro stays the choice when a brief needs the family's highest factual accuracy. If 2.1 and Pro both fit a job, run the same brief through both and compare the outputs and the quotes you are shown.
What to compare
Nano Banana 2.1
Nano Banana 2
Nano Banana 2 Lite
Google's positioning
Flash-tier generation and editing, the generally available successor to Nano Banana 2
Fastest and most cost-efficient tier in the family, built for throughput
Announced
October 2026
February 2026
June 2026
Model ID
gemini-nano-banana-2.1
gemini-3.1-flash-image-preview
gemini-3.1-flash-lite-image
Resolution here
1K, 2K, 4K
1K, 2K, 4K
1K only
Aspect ratios here
14 fixed values, no auto
15 values, including auto
7 fixed values, no auto
Reference images here
Up to 14
Up to 14
Up to 5
Pick it when
You want the current Flash-tier model with the newest text and consistency work
You want fast iteration with groundable detail and legible text
You need many cheap 1K frames for ideation
Editing with references instead of masks
An edit starts with images you are allowed to use and an instruction that says what should change and what must survive it. There is no mask to draw: the model reasons about which parts the instruction refers to and leaves the rest alone.
Write the instruction in two halves. First the object and the change. Then the parts that must not move, such as the silhouette, the camera angle or the label on the packaging.
Check faces and hands first. Resemblance drift is the most common unwanted change in a portrait edit.
Re-read small text. Lettering, prices and legal wording are the second most common casualty.
Compare reflections, shadows and contact points where an object was added or removed.
Change one instruction at a time between runs, so you can attribute the difference to the instruction and not to the model's variance.
How to generate or edit here
Four steps, in the order the workspace asks for them.
Paste your prompt, or open a prompt from the section above with the copy button.
Choose a resolution tier and an aspect ratio. Add reference images when the job is an edit rather than a new image.
Read the quote, then sign in when the workspace asks for it.
Submit, review the result against the check list below, and download the version you want to keep. If a task fails, read its status before retrying so you do not submit a duplicate.
Writing a prompt that survives revision
The guidance that works for this model family is a formula rather than a keyword list. Name the deliverable, then the subject, the action and the scene, and build from there.
Be specific rather than short. A weak instruction names a subject; a stronger one adds what the subject is doing, where, and in what style. Quote any text that must appear verbatim, decide the composition, the style and the aspect ratio before you generate, and write the ratio into the prompt when the tool you use accepts it.
When you revise, change one decision at a time. A prompt that edits four things at once produces a result you cannot attribute.
Output check list
Run this list over every result before you download or publish it, especially in the first weeks of a new model, when reports about what it does well and what it still misses are mostly individual experiences.
Faces and hands. Resemblance to a reference is a tendency, not a guarantee; check identity-critical features first.
Text. Lettering, numbers, prices and legal wording at full size; shorten and re-quote anything garbled.
Spatial relations. Left and right can swap in complex scenes; check anything directional.
Consistency across a set. Compare every frame against the reference set, not only against the previous frame.
Pose and framing. An edit can keep the original pose when you asked for a new one; say explicitly when the pose should change.
Provenance. Outputs carry C2PA Content Credentials per Google; keep that in mind before stripping metadata.
Early impressions. The first days of user discussions mention prompt adherence and resemblance as the points worth double-checking. Treat those as individual experiences rather than verified conclusions, and let your own outputs decide.
When a task fails or disappoints
Most failures have a cheap recovery. The list below covers the ones you will actually meet in this workspace.
Task failed. Read the status message before retrying, so you do not submit a duplicate charge for the same brief.
Service temporarily unavailable. The catalog marks the mode unavailable while the upstream is down; wait and retry, or run the same brief on Nano Banana 2 or Nano Banana Pro meanwhile.
Instruction ignored. Split a long instruction into one change per run, and name the parts that must stay unchanged.
Garbled text. Shorten the quoted text, fix its spelling in the prompt, and move to a higher resolution tier.
Drifting resemblance. Re-attach the full reference set and restate the identity-critical features by name.
Not enough balance. The quote shown before submission is the binding figure; top up from the pricing page before a longer run.
Case coverage this page will validate
The gallery above shows what the model is pitched for, with third-party material, while this site's own runs are still pending. The five task types below stay the case cards this page will fill first, each with a starter prompt in the section above. None of them becomes a published capability until real runs on this service back it.
When the runs exist, each card will carry its prompt, its source and its provenance, and the image will be generated on this site rather than borrowed from a third party.
Posters and social media visuals with a headline.
Product photography in a styled scene.
Reference-based edits of an uploaded photo.
Layouts with rendered text such as menus and cards.
Portrait and character consistency across a set of images.
Where the facts come from
Every figure on this page traces back to one of the sources below, and the table says what each one is allowed to support.
Source
Used for
Reference
Google Cloud documentation for Nano Banana 2.1, retrieved 7 October 2026
Model ID, release status, limits, capabilities and unsupported features
It is Google's image generation and editing model, released as a generally available version on 6 October 2026 under the model ID gemini-nano-banana-2.1. Google's documentation positions it in the Flash tier as the successor to Nano Banana 2, with Nano Banana Pro remaining the family's higher-fidelity tier.
Can I use Nano Banana 2.1 for free here?
This page promises no free generations and no signup allowance. Check your account balance and read the quote the workspace shows before you submit. Whatever credits and payment options exist are listed in your account and on the pricing page.
Which modes are open on this site?
Text-to-image and image-to-image, per the live service catalog retrieved on 7 October 2026. Google's documentation mentions further capabilities, such as images derived from video input, but this service does not open them, and this page does not promise them.
How do I edit a photo instead of generating a new image?
Upload the reference, then describe the change and the parts that must stay. State the object and the adjustment in one sentence, and add a second sentence naming what should not move. Compare the result with the source for faces, proportions, reflections and small text.
Will the same character survive across several images?
Google documents an input limit of up to 14 reference images per prompt, which is what makes multi-image consistency possible. Keep the reference set attached to every request and name the character each time. Consistency is a tendency rather than a guarantee: early user discussions cite it as the most noticeable improvement area of this release, but that is an individual experience, not a verified conclusion, so review each frame yourself.
Can it render readable text inside an image?
Legible lettering is a documented strength of this model family, and the starter prompts above include a layout task that exercises it. Quote the exact wording in the prompt and proofread anything critical at full size, because small text is also one of the first things to check when a result disappoints.
How is Nano Banana 2.1 different from Nano Banana 2?
Google positions 2.1 as the generally available successor in the same Flash tier. In this workspace both expose 1K, 2K and 4K, 14 or 15 aspect ratio values and up to 14 reference images, so the practical difference is the newer consistency and text work rather than a bigger ceiling. Run the same brief through both when the choice matters.
And how is it different from Nano Banana Pro?
Pro remains the family's higher-fidelity tier for briefs that need maximum factual accuracy, while 2.1 is the Flash-tier model for fast generation and editing. Early user impressions that the outputs look close to Pro are individual experiences rather than verified results; compare on your own briefs before deciding.
What about Nano Banana 2 Lite?
It is the family's fastest and most cost-efficient tier, announced on 30 June 2026 under the model ID gemini-3.1-flash-lite-image. In this workspace it is fixed at 1K with up to 5 reference images and 7 aspect ratios, which makes it the budget choice for ideation frames.
Which resolutions and aspect ratios can I choose?
This workspace offers 1K, 2K and 4K with 1K as the default, and 14 fixed aspect ratios from 1:1 to 21:9, including 9:16, 16:9 and the extreme 1:4, 4:1, 1:8 and 8:1 frames. There is no auto value. Other sites list a different set, and their list does not apply to this account.
Why is there no seed or temperature control?
Because Google rejects them for this model. The official documentation states that seed, temperature, topP and topK are not supported and that passing them returns an error, so this workspace does not offer them. Reproducibility has to come from fixed references and wording instead.
Does the output carry a provenance marker?
Google documents C2PA Content Credentials for this model. Nothing on this site removes provenance metadata, and the check list above reminds you to keep that in mind before stripping metadata yourself.
What does a run cost?
The workspace quote is the only number that applies to your task, and it is resolved by the service before submission. The service catalog retrieved on 7 October 2026 starts at 3 credits per result at 1K, with 4 credits at 2K and 6 credits at 4K. Third-party rate cards, such as fal's 0.08 USD per image at 1K, describe those providers' own accounts.
Explore details ↗ 
